CN101916447A - Robust motion target detecting and tracking image processing system - Google Patents

Abstract

The invention provides a robust motion target detecting and tracking image processing system. The system comprises a digital signal processor taking DSP as a kernel, a CCD camera and a PC, and executes the following steps of: acquiring video data; reading acquired digital video sequences frame by frame by using the digital signal processor; performing motion target detection on the frame sequences in the last step by adopting an improved mixture Gaussian background model; performing removing noise and shadow from the detection result in the last step; tracking the detected motion target; and transmitting and displaying the processed result to a user for processing, wherein adoption of the improved mixture Gaussian background model comprises the following steps of: performing background model initialization on the input video frame to establish a background model; subtracting the current frame from a background picture; binaryzing a foreground picture; morphologically filtering to remove the noise; introducing an accelerating factor and a smooth equation to update the background model; and converting the background into an image format for display.

Description

A kind of moving object detection and tracking image processing system of robust

Technical field

The invention belongs to the digital image processing techniques field, be specifically related to a kind of moving object detection and tracking image processing system of robust, realize abnormal motion target in the vision signal of input is carried out detection and tracking.

Background technology

DSP (Digital Signal Processor) is a kind of microprocessor, handles the device of a large amount of digital signals.It is not only programmable, and travelling speed can reach the per second number with ten million bar complicated order program in fact the time, considerably beyond general purpose microprocessor, is the chip that becomes more and more important in the digitalized electron world.Big, real-time, the high requirement of message transmission rate of operand in the Flame Image Process because DSP is content with very little, furthermore computing machine strong multimedia interaction capabilities, thereby DSP is widely used in image processing field.

Moving object detection is meant from sequence image region of variation is split from background.It is a major part of digital image processing techniques, be the emphasis and the difficult point of research fields such as computer vision, target recognition and tracking, pattern-recognition, moving image encoding, content-based retrieval, security monitoring, have a wide range of applications in fields such as traffic, national defence and industry.Yet,, bring very big challenge for the correct detection of moving target owing to the existence of the reasons such as motion of the shadow of the interference of the variation of weather and illumination, background clutter motion, moving target and video camera.Because can the correct detection of moving target affect moving target and be followed the tracks of and classification by correct with cutting apart, and therefore become an important problem in the video monitoring system research.Its basic task is to detect movable information from image sequence, and the simplified image processing procedure obtains required motion vector, thereby can the recognition and tracking object.

Moving target detecting method commonly used at present has three kinds: optical flow method, adjacent frame difference method, background subtraction.The present invention be directed to moving object detection in the computer intelligence video monitoring system.A typical video monitoring system is to monitor fixed area in real time with the video camera of a static state, its objective is from static background, to be partitioned into dynamic object, and to its classify, operation such as tracking.So cutting apart dynamic object from video flowing in real time is basic link of video monitoring system.Moving object detection is in the bottom of whole computer intelligence video monitoring system, is the basis of various follow-up advanced processes such as target following, target classification, behavior understanding.For still camera, background modeling is to solve the effective ways of cutting apart dynamic object in real time.

An effective background model should be able to overcome following several problem that usually exists in actual applications:

(1) extraction of background model: the simplest method of obtaining of background image is not carry out under scene has the situation of moving target, but can't satisfy this requirement in some application scenario.

(2) disturbance of background: can contain the object of slight disturbance in the background, as shaking of branch, leaf, the disturbance part should not be counted as the foreground moving target.

(3) variation of extraneous light: in one day the variation of different time sections light, weather etc. and indoor turn on light, turn off the light etc. all can exert an influence to testing result.

(4) renewal of background:, need in time background model to be upgraded, to adapt to this variation because the variation of illumination or other condition can make background image change.

(5) influence of shade: the shade of foreground target is also detected as the part of moving target usually, will influence further processing and analysis to moving target like this.

The present invention sums up on the basis of characteristics of existing algorithm, makes improvements, and has finally provided one and can adapt to the slight disturbance of background scene, overcomes noise and ambient lighting is influence, object detection system real-time update, robust.

In the research field of computer vision, the sequence image motion target tracking is an ancient and not fine yet solution, challenging important topic.Target following has caused people's very big concern owing to have a wide range of applications and demand.It mainly contains industrial process in the application aspect civilian and controls, medical research, and the industrial products supervision, traffic monitoring, image retrieval and recovery, the video coding of based target, compression reach synthesizes or the like.By comparison, multiple target tracking application militarily is subjected to various countries scholar's attention more.Target following mainly contains battlefield surveillance in the application of military aspect, and military target is followed the tracks of, sea survaillance, air-borne early warning etc.In different applications, be different to the requirement of motion target tracking.As in the video of based target coding and video synthesize, precision is very important; And in video monitoring, require to follow the tracks of automatically real-time, can tolerate certain error.

The present invention is that the multiple mobile object in the exemplary video supervisory system is followed the tracks of, and the work of target following is to be based upon on the basis of moving object detection to finish.

Summary of the invention

The main application scenario of the present invention is that outdoor sport target detection and indoor abnormal behaviour detect and follows the tracks of.This system has adopted the software coding method of robust, flexible design, and man machine operation interface is comparatively flexibly adopted in the modification of algorithm and optimize conveniently, easy to use, has engineering range of application widely.

The technical scheme that realizes the object of the invention is: a kind of moving object detection and tracking image processing system of robust, comprise with DSP being digital signal processor, ccd video camera and the PC of core, one end of digital signal processor connects ccd video camera, the other end connects PC, and described image processing system is carried out following steps:

The collection of step 1 video data: digital signal processor, is transferred in the memory buffer space of appointment behind the analog to digital converter digital collection from the analog video signal of camera CCD input, waits pending;

Step 2 digital signal processor reads in the digital video sequences that collects frame by frame, adopts improvement mixed Gaussian background model that the frame sequence in the step 2 is carried out moving object detection;

Step 3 is carried out denoising and shadow removal with the testing result in the step 2;

Step 4 is followed the tracks of detected moving target on the basis of step 3;

Step 5 transmits result and is shown to the user, with pending;

Wherein, the improved mixed Gaussian background model of described employing comprises the following steps:

The frame of video of step 21 pair input is carried out the initialization of background model, sets up background model;

Step 22 present frame subtracts each other with Background;

Step 23 binaryzation foreground picture;

Step 24 is carried out morphologic filtering, removes noise;

Step 25 is introduced speedup factor and smoothing equation comes update background module;

Step 26 is converted into picture format with background, in order to show.

Initialization background model in the described step 21 specifically comprises:

Each pixel is set up histogram that one section samples in the time and histogram is carried out Filtering Processing, obtain the position at spike place by histogrammic first order difference, to each spike the parameter of corresponding Gaussian distribution carry out initialization.

The renewal of background model specifically comprises in the described step 25:

Definition θ _{X, y, i}(t)=t ²Be speedup factor, the pixel that x wherein, the coordinate of pixel in the y representative image, t represent to be positioned at this position drops on i the duration on distributing, when distributing coupling for j in pixel that newly obtains and the Gauss model, then j weight that distributes upgraded by following equation:

Wherein α is some fixing constants.

In the background model renewal process, introduce speedup factor and rationality and judge, can the acceleration model updating efficiency, produce the smear phenomenon when effectively solving static target and moving.

When not changing for a long time, as long as scene illumination has slight change just to be detected.Problem hereto, the present invention adopts smoothing equation that K Gaussian distribution upgraded.

$w_{t,v,j,n+1}=\frac{1+w_{v,j,n}\mathrm{\α}/{\mathrm{\θ}}_{x,y,j}\left(t\right)}{\mathrm{\α}/{\mathrm{\θ}}_{x,y,j}\left(t\right)+1}$

$f_{n+1}=\left\{\begin{array}{c}{f}_n+1\mathrm{if}{l}_n\≠255\\ f_n\mathrm{if}{l}_n=255\end{array}\right.$

l _n+1＝255

W wherein _{V, k, n}, f _n, l _nBe respectively n weights, recurrent frequency and the life span of n the model in pixel x place constantly.

Description of drawings

Fig. 1 video data acquiring flow process.

Fig. 2 moving object detection process flow diagram.

Fig. 3 shadow removal process flow diagram.

Fig. 4 motion target tracking process flow diagram.

Fig. 5 is that embodiment of the invention image processing system is carried out schematic flow sheet

Embodiment

Further specify below in conjunction with embodiment and accompanying drawing.

A kind of moving object detection and tracking image processing system of robust, comprise with DSP being digital signal processor, ccd video camera and the PC of core, one end of digital signal processor connects ccd video camera, and the other end connects PC, and image processing system is carried out following steps:

The collection of step 1 video data: digital signal processor, is transferred in the memory buffer space of appointment behind the analog to digital converter digital collection from the analog video signal of camera CCD input, waits pending;

Step 2 digital signal processor reads in the digital video sequences that collects frame by frame, adopts improvement mixed Gaussian background model that the frame sequence in the step 2 is carried out moving object detection;

Step 3 is carried out denoising and shadow removal with the testing result in the step 2;

Step 4 is followed the tracks of detected moving target on the basis of step 3;

Step 5 transmits result and is shown to the user, with pending.

As shown in Figure 1, the collecting flowchart of step 1 video data comprises the following steps:

Step S101 begins video acquisition by the CCD camera;

Step S102 opens the CCD camera;

Step S103 carries out analog to digital conversion (A/D) with video data;

The video requency frame data that step S104 changes is to dsp processor application frame cushion space;

The frame buffered video that step S105 will apply for is all as formation;

Step S106 begins the collection of frame of video.

As shown in Figure 2, step 2 adopts improved mixed Gaussian background model to carry out moving object detection to comprise the following steps

Step S201 carries out the initialization of background model to the frame of video of input, sets up background model;

Step S202 present frame subtracts each other with Background;

Step S203 binaryzation foreground picture;

Step S204 carries out medium filtering and morphologic filtering, removes noise;

Step S205 introduces speedup factor and smoothing equation comes update background module;

Step S206 is converted into picture format with background, in order to show.

Initialization background model among the step S201 wherein specifically comprises:

Each pixel is set up histogram that one section samples in the time and histogram is carried out Filtering Processing, obtain the position at spike place by histogrammic first order difference, to each spike the parameter of corresponding Gaussian distribution carry out initialization.

Wherein the renewal of background model specifically comprises among the step S205:

Definition θ _{X, y, i}(t)=t ²Be speedup factor, the pixel that x wherein, the coordinate of pixel in the y representative image, t represent to be positioned at this position drops on i the duration on distributing, when distributing coupling for j in pixel that newly obtains and the Gauss model, then j weight that distributes upgraded by following equation:

Wherein α is some fixing constants.

In the background model renewal process, introduce speedup factor and rationality and judge, can the acceleration model updating efficiency, produce the smear phenomenon when effectively solving static target and moving.

When not changing for a long time, as long as scene illumination has slight change just to be detected.Problem hereto, the present invention adopts smoothing equation that K Gaussian distribution upgraded.

$w_{t,v,j,n+1}=\frac{1+w_{v,j,n}\mathrm{\α}/{\mathrm{\θ}}_{x,y,j}\left(t\right)}{\mathrm{\α}/{\mathrm{\θ}}_{x,y,j}\left(t\right)+1}$

$f_{n+1}=\left\{\begin{array}{c}{f}_n+1\mathrm{if}{l}_n\≠255\\ f_n\mathrm{if}{l}_n=255\end{array}\right.$

l _n+1＝255

W wherein _{V, k, n}, f _n, l _nBe respectively n weights, recurrent frequency and the life span of n the model in pixel x place constantly.

As shown in Figure 3, testing result is carried out denoising and shadow removal comprises the following steps:

Step S301 reads the prospect frame;

Step S302 calculates S ^R, S ^GAnd S ^B, and get maximal value and be designated as S;

With the corresponding pixel component value of foreground pixel component maximal value, be designated as S ' in the step S303 calculating background, and judgement S (i, j)-S ' (i, j)＜T, wherein T is a threshold value, if set up execution in step S304, if be false execution in step S305;

Step S304 detects shade and removal;

Step S305 obtains removing the prospect frame behind the shade.

As shown in Figure 4, carrying out motion target tracking according to testing result comprises the following steps:

Step S401 takes out the moving region according to testing result;

Step S402 initialization searching times threshold value;

Step S403 is according to the current location of Kalman filtering method target of prediction;

Step S404 is according to Mean-Shift ferret out zone, and the searching times threshold value subtracts one;

Step S405 judges whether searching times is successful, successfully goes to step 46, and failure goes to step 47;

Step S406 uses the successful result of search, goes to step 43;

Step S407 judges whether to reach searching threshold, goes to step 48 if reach, if do not go to step 43;

Search is accepted in the out-of-bounds of step S408 target.

As shown in Figure 5, the specific implementation flow process of image processing system in hardware.

B1: after the data of camera collection deposit input buffering in, from input buffering, read in the image of camera collection;

B2:, and carry out pre-service to image coding and decoding;

B3: pretreated image information buffer memory to dsp chip, is carried out the processing of moving object detection and tracking;

B4: the image information after the dsp chip processing is transmitted by Internet;

B5: the result that dsp chip is handled shows the desktop to the user.

Claims (5)

1. the moving object detection and tracking image processing system of a robust, comprise with DSP being digital signal processor, ccd video camera and the PC of core, one end of digital signal processor connects ccd video camera, and the other end connects PC, and image processing system is carried out following steps:

The collection of step 1 video data: digital signal processor, is transferred in the memory buffer space of appointment behind the analog to digital converter digital collection from the analog video signal of camera CCD input, waits pending;

Step 2 digital signal processor reads in the digital video sequences that collects frame by frame, adopts improvement mixed Gaussian background model that the frame sequence that collects is carried out moving object detection;

Step 3 is carried out denoising and shadow removal with the testing result in the step 2;

Step 4 is followed the tracks of detected moving target on the basis of step 3;

Step 5 transmits result and is shown to the user, with pending;

It is characterized in that the improved mixed Gaussian background model of described employing comprises the following steps:

The frame of video of step 21 pair input is carried out the initialization of background model, sets up background model;

Step 22 present frame subtracts each other with Background;

Step 23 binaryzation foreground picture;

Step 24 is carried out morphologic filtering, removes noise;

Step 25 is introduced speedup factor and smoothing equation comes update background module;

Step 26 is converted into picture format with background, in order to show.

2. the moving object detection and tracking image processing system of robust according to claim 1, it is characterized in that, initialization background model in the step 21, specifically comprise: each pixel is set up histogram that one section samples in the time and histogram is carried out Filtering Processing, obtain the position at spike place by histogrammic first order difference, to each spike the parameter of corresponding Gaussian distribution carry out initialization.

3. the moving object detection and tracking image processing system of robust according to claim 1 is characterized in that, the renewal of background model specifically comprises in the step 25:

Definition θ _{X, y, i}(t)=t ²Be speedup factor, the pixel that x wherein, the coordinate of pixel in the y representative image, t represent to be positioned at this position drops on i the duration on distributing, when distributing coupling for j in pixel that newly obtains and the Gauss model, then j weight that distributes upgraded by following equation:

Wherein α is some fixing constants; Adopt smoothing equation that K Gaussian distribution upgraded,

$w_{t,v,j,n+1}=\frac{1+w_{v,j,n}\mathrm{\α}/{\mathrm{\θ}}_{x,y,j}\left(t\right)}{\mathrm{\α}/{\mathrm{\θ}}_{x,y,j}\left(t\right)+1}$

$f_{n+1}=\left\{\begin{array}{c}{f}_n+1\mathrm{if}{l}_n\≠255\\ f_n\mathrm{if}{l}_n=255\end{array}\right.$

l _n+1＝255

W wherein _{V, k, n}, f _n, l _nBe respectively n weights, recurrent frequency and the life span of n the model in pixel x place constantly.

4. the moving object detection and tracking image processing system of robust according to claim 1 is characterized in that, described step 3 pair testing result carries out denoising and shadow removal comprises the following steps:

Step S301 reads the prospect frame;

Step S302 calculates S ^R, S ^GAnd S ^B, and get maximal value and be designated as S;

With the corresponding pixel component value of foreground pixel component maximal value, be designated as S ' in the step S303 calculating background, and judgement S (i, j)-S ' (i, j)＜T, wherein T is a threshold value, if set up execution in step S304, if be false execution in step S305;

Step S304 detects shade and removal;

Step S305 obtains removing the prospect frame behind the shade.

5. the moving object detection and tracking image processing system of robust according to claim 1 is characterized in that, described step 4 is carried out motion target tracking according to testing result and comprised the following steps:

Step S401 takes out the moving region according to testing result;

Step S402 initialization searching times threshold value;

Step S403 is according to the current location of Kalman filtering method target of prediction;

Step S404 is according to Mean-Shift ferret out zone, and the searching times threshold value subtracts one;

Step S405 judges whether searching times is successful, successfully goes to step 46, and failure goes to step 47;

Step S406 uses the successful result of search, goes to step 43;

Step S407 judges whether to reach searching threshold, goes to step 48 if reach, if do not go to step 43;

Search is accepted in the out-of-bounds of step S408 target.

Images